Greetings! This is Amit Pandey. A Researcher with Passion for Products, Poetry and Teaching.

I am a Data Science Researcher working with Prof. Vikram Pudi at the Data Sciences and Analytics Center, KCIS-IIIT Hyderabad. My interests include Statistical Al, Applied Machine Learning, Natural Language Processing & Understanding, Computer Vision, and Business Intelligence. Experienced Product Manager with a demonstrated history of working in the fast-paced tech startup environment.

Currently working on: Feature-rich contextual embeddings for downward tasks such as Classification, Ranking, and Recommendation.

Reach out for discussing new ideas and for research collaboration.

Education

• International Institute of Information Technology, Hyderabad, India (2020-Present) : MS by Research - focused on Statistical AI, Applied NLP and Multimodal Machine Learning. (9 CGPA)
• BVCOE, GGSIPU, New Delhi, India : B.Tech in ECE - First Division (8.07 CPGA)
• Intermediate - CBSE : PCM with Computer Science - 92.2%

Related Courses

• Statistical Methods in AI , Optimization Methods, Advanced NLP, Applied NLP, Computer Vision, Digital Image Processing, Topics in Applied Optimization.

Work experience

• Aug 2020- Present: Data Science Researcher
  • DSAC, IIIT Hyderabad
  • Supervisor: Professor Vikram Pudi
  • Currently working on developing feature-rich embeddings for a wide range of downward tasks such as Classification, Ranking, and Recommendation. Also exploring Text Summarization, Creative NER and Explainability in AI.
  • The current role focuses on exploring and optimizing Applied Machine Learning solutions.
• Oct 2021-Present: Lab Instructor and Teaching Assistant
  • FMML, IIIT Hyderabad
  • Supervisor: Professor C V Jawahar and Anoop Namboodiri
  • Lab instructor and teaching assistant for UG and PG students enrolled in Foundations of Modern Machine Learning course. Also take office hours and doubt clearing sessions.
• May 2018 - Aug 2020 : Product Manager
  • Aphelia Innovations
  • Optimized Products and services to best suit clients requirements by working closely on all the parts of the Product lifecycle- Conceptualisation, Designing, Testing, and Deployment and Marketing. Ensured on-time delivery of high-quality products and services by leading a team of programmers, designers, and marketing professionals. Created a hassle-free experience for clients with effective communication

Projects

• Blue Whale detection
  • Developed a system to detect and identify Blue whales (endangered species). A Light, fast and Robust system giving an accuracy comparable to the system developed by Stanford University team. (To be submitted )
• Biometric Pattern Classification for Fasater Identification
  • Created a pipeline to classify finger print patters for downward task of Identification. It acted as preliminary step for reducing complexity of database search in the problem of automatic fingerprint matching . Though it didn’t out perform neural networks, it explored SVM’s capability of learning sparse high dimensional data. Low training data requirement, introduction of Error of Correcting Codes paired with carefully picked image processing techniques gave impressive results. Improved speed of identification by more than 2x and accuracy by around 7% (over other non-dl techniques).
• Scientific Document Summarization
  • Summarization of scientific articles is a widely studied problem. Incorporating the additional information like facets of research information along with the summaries of papers written by other papers citing it can be useful for the task of summarization. Citations contain meta-commentary and provide additional contextual information about a reference paper. Since the citances contain some specific excerpts of the paper depending on their use-case, identifying which exert in reference paper a particular citance is referring to is helpful. Thus defining the summarization task of the reference paper based on its citation papers along with using an abstract of the reference paper helps by providing additional contextual information. Developed pipeline for better recall of RPs for given Citances of CPs. (To be submitted)
• Feature rich word embeddings : (Flock of Words)
  • At the core of any NLP task are the word embeddings which are able to capture the details of the language in higher dimensional space. From classical to modern machine learning methods, all benefit from finer pre-trained word embeddings that can reduce overhead at the time of computation and improve accuracy. Improved accuracy on standard NLP tasks such as classification and STS by 10% (relative to the baseline). (To be submitted)
• GrabCut
  • Implemented GrabCut(Microsoft)- An Interactive Foreground Extraction using Iterated Graph Cuts based on Maxflow-Mincut optimization problem.
• Panaroma Generator
  • Stitch together multiple images of same scene (in random order) with some overlapping portion to get a panaromic image. Can be used for indoor and outdoor scene recreation.
• Suicidal Tendency detection
  • Conceptualizing RNN/Bi-LSTM/Transformer based sentiment analysis specifically to detect suicidal tendencies in a person.
• Multilingual NER detection in creative work
  • Working on solving NER problem in artistic/creative works such as poems and speeches.